[dbnode] Remove allocation per series ID when streaming block from disk

src/dbnode/network/server/tchannelthrift/node/service.go

@@ -808,12 +808,14 @@ func (s *service) fetchReadEncoded(ctx context.Context,
 	// Re-use reader and id for more memory-efficient processing of
 	// tags from doc.Metadata
 	reader := docs.NewEncodedDocumentReader()
-	id := ident.NewReusableBytesID()
 	for _, entry := range results.Map().Iter() {
 		idx := i
 		i++
 
-		id.Reset(entry.Key())
+		// NB(r): Use a bytes ID here so that this ID doesn't need to be
+		// copied by the blockRetriever in the streamRequest method when
+		// it checks if the ID is finalizeable or not with IsNoFinalize.
+		id := ident.BytesID(entry.Key())
 
 		d := entry.Value()
 		metadata, err := docs.MetadataFromDocument(d, reader)

src/dbnode/persist/fs/retriever.go

@@ -565,9 +565,17 @@ func (r *blockRetriever) streamRequest(
 	nsCtx namespace.Context,
 ) (bool, error) {
 	req.shard = shard
-	// NB(r): Clone the ID as we're not positive it will stay valid throughout
-	// the lifecycle of the async request.
-	req.id = r.idPool.Clone(id)
+
+	// NB(r): If the ID is a ident.BytesID then we can just hold
+	// onto this ID.
+	seriesID := id
+	if !seriesID.IsNoFinalize() {
+		// NB(r): Clone the ID as we're not positive it will stay valid throughout
+		// the lifecycle of the async request.
+		seriesID = r.idPool.Clone(id)
+	}
+
+	req.id = seriesID
 	req.start = startTime
 	req.blockSize = r.blockSize